Credal networks generalize Bayesian networks relaxing numerical parameters. This considerably expands expressivity. but makes belief updating a hard task even on polytrees. Nevertheless, if all the variables are binary, ...